美股

On September 8, 2026, 100% of Foreign Net Buying in Korea Went to Three Memory Names — So Why Did Samsung Electronics Still Close Down 0.19%? A Price-Efficiency Decomposition

2026-09-08 · 11 min · AlphaGBM
foreign investor flowsKRXKorea ExchangeSamsung Electronics005930SK Hynix000660Hanmi Semiconductor042700memory semiconductorsHBMDRAMNANDsemiconductor equipmentprice efficiencyflow decompositiondistribution vs accumulationforeign ownership ratiozero-sum investor accountingretail absorptionsector rotationMicronMUSanDiskSNDKWestern DigitalWDCSMHKorea equitiesAsia semiconductorsfalsification criteriamarket microstructureinstitutional flows2026-09-08外资流向韩国交易所三星电子SK海力士韩美半导体存储半导体半导体设备推价效率派发结构外资持股比板块轮动资金流拆解可证伪判据

Data basis (read this first)

  • Korean figures: settled closing values from the Korea Exchange (KRX) session of 2026-09-08. These are the sole basis for this article's conclusions.
  • U.S. figures: ⚠️ U.S. markets were still open at the time of writing (the 09:30–16:00 ET session had not closed). All U.S. numbers below are intraday prints, as-of 10:46 ET on 2026-09-08, measured against the 2026-09-04 settled close (2026-09-07 was Labor Day, markets closed). Intraday prints change continuously; no U.S. figure is used in this article's title or conclusion.
  • Hong Kong figures: not referenced; the session had not entered our readings layer.
  • FX: live KRW/USD, as-of 2026-09-08. USD values are conversions for order-of-magnitude reference only; all calculations use native KRW.
  • This is public-layer research containing only public market data, methodology, and opinion. It contains no position, allocation, or trade-execution information.

Bottom line

Foreign investors' sector selection in Korea became more extreme, not less — 100% of gross foreign buying went to three memory-chain names. But three things happened simultaneously: total size contracted, the concentration point migrated from chips toward equipment, and the lead name lost its ability to move price. Directional concentration alone cannot be read as a bullish signal.

1. Breadth vs. weight: the prior session's test came due today

The prior session (09-07) read as "broad by count, but 97% of the money went to three names." The pre-registered test was: if non-semiconductor names flip back to net selling, then what is happening is selective buying, not buying Korea. That test came due today, and it resolved cleanly.

Metric (8-name sample) 09-07 close 09-08 close Change
Names closing higher 7 / 8 2 / 8 −5
Foreign net-buy breadth 6 buy / 2 sell 3 buy / 5 sell Flipped
Gross foreign buying (close-weighted) KRW 2.341tn KRW 1.125tn −52.0%
Net foreign flow (close-weighted) KRW 2.298tn KRW 1.026tn (~USD 0.76bn) −55.4%
Memory chain as % of gross buying 97.11% 100.00% +2.89pp

The buy and sell sides split along the semiconductor / non-semiconductor line with zero crossovers:

Eight names, zero exceptions. This is not coincidence at the sample size; it is a directional rotation.

2. The novel finding: an order-of-magnitude gap in price efficiency

This is the core of the analysis and the layer generally missing from public commentary. Among memory names all receiving foreign net buying, each unit of capital bought a completely different amount of price.

Name Foreign net (shares) Institutional net (shares) Combined as % of volume Day's move
Samsung Electronics +3,332,528 +2,406,250 25.4% −0.19%
SK Hynix +97,545 −14,560 2.0% +0.56%
Hanmi Semiconductor +212,271 +93,889 38.1% +1.66%

Samsung Electronics traded 22,563,172 shares. Foreign and institutional investors together net-bought 5,738,778 of them — a quarter of the entire day's volume — and the stock still closed lower. The same day, SK Hynix closed up 0.56% on a combined net-buy ratio of just 2.0%.

Applying the zero-sum identity of KRX investor-type accounting (foreign + institutional + retail/other = 0):

Counterparty to Samsung Electronics = retail and other, net selling 5,738,778 shares

In other words, the institutional bid was absorbed in full by retail supply, leaving no residual bid to lift the price. This is a distribution structure. In flow data it looks identical in direction to accumulation; in price it produces the opposite outcome. Reading net-buy direction alone conflates the two.

3. Size is shrinking; the concentration point is migrating from chips to equipment

While sector concentration rose, the weighting within the three memory names was reshuffled dramatically:

Name % of gross buying 09-07 % of gross buying 09-08 Change in foreign net shares
SK Hynix 58.79% 15.55% 772,036 → 97,545 (−87.4%)
Samsung Electronics 37.44% 79.84% 3,246,247 → 3,332,528 (+2.7%)
Hanmi Semiconductor 0.88% 4.61% 86,155 → 212,271 (+146.4%)

SK Hynix, the prior session's lead recipient, saw foreign buying shrink 87.4% and institutions flip to net selling (−14,560 shares). The most concentrated pressure moved to Hanmi Semiconductor on the equipment side, at 38.1% of its own daily volume — the highest of the three.

4. The slowest-moving variable confirms the same direction

Net-bought share counts are a same-day flow and can be distorted by a single block trade. Foreign ownership ratio is a stock measure — the slowest to move and the hardest for one-day noise to contaminate. Across all eight names, the direction of the ownership ratio matched the direction of that day's net flow, one for one, with no exceptions:

Name Foreign ownership 09-07 → 09-08 Change
Hanmi Semiconductor 8.33% → 8.56% +0.23pp (largest in sample)
Samsung Electronics 46.79% → 46.85% +0.06pp
SK Hynix 50.63% → 50.65% +0.02pp
Kia 39.24% → 39.23% −0.01pp
LG Energy Solution 5.53% → 5.51% −0.02pp
Samsung SDI 25.51% → 25.49% −0.02pp
LG Electronics 29.72% → 29.68% −0.04pp
Hyundai Motor 24.43% → 24.36% −0.07pp

All three semiconductors up, all five non-semiconductors down. Flow and stock agree, so this rotation is real rather than one-day noise.

5. External validation has not yet arrived (U.S. markets still open)

When these two Korean sessions completed (09-07 sharply higher, 09-08 divided), of the other two markets one had not opened and one was falling. This repricing is currently one-sided and externally unvalidated.

Two-session cumulative from the 09-04 close (Korea, settled):

Name 09-04 close 09-08 close Two-session cumulative
SK Hynix KRW 1,647,000 KRW 1,793,000 +8.86%
Hanmi Semiconductor KRW 230,000 KRW 244,500 +6.30%
Samsung Electronics KRW 255,500 KRW 269,500 +5.48%

⚠️ The following are intraday prints, as-of 2026-09-08 10:46 ET. The session had not closed, values change continuously, and these do not inform this article's conclusion (measured against the 09-04 close):

Ticker Intraday move (vs 09-04 close)
SNDK (SanDisk) +2.60%
WDC (Western Digital) +2.29%
SMH (semiconductor ETF) +1.38%
MU (Micron) +0.35%
QQQ (Nasdaq-100 ETF) −0.11%

Until this session closes, "sector repricing" is not a conclusion that can be drawn. Worth noting: MU was sampled three times while this article was being written, across a span of 14 minutes, and changed sign twice: +0.09% → −0.13% → +0.35% — direct evidence of why an intraday print cannot serve as a conclusion. Real validation occurs at the 16:00 ET close, when a comparable settled value first exists.

6. Falsification criteria (full-range, computed at the next close)

Criteria are specified in relative magnitudes and continuous bands, so that no outcome falls into an undefined gap between endpoints.

Test 1 · Is the broadening real? (combined foreign net flow of the 7 names excluding Samsung)

Next session, 7-name combined net Reading
≥ +1.0mn shares Broadening confirmed; back to 09-07-style market-wide re-risking
0 to +1.0mn shares Marginal improvement, but still structural selection
−0.5mn to 0 shares Status quo (09-08 was −27,804 shares)
≤ −0.5mn shares Selection intensifying; active de-risking on the non-semiconductor side

Test 2 · Does Samsung's distribution persist? (combined net-buy % of volume × price direction)

Combined % of volume Price Reading
≥ 20% Down Distribution persists; bid still being absorbed (09-08 was 25.4% / −0.19%)
≥ 20% Up Supply exhausted; broadening finally corroborated by price
10%–20% Either Institutional participation declining; this test loses information value
< 10% Either Institutions have left; the distribution thesis is void

Test 3 · Does the migration persist? (Hanmi Semiconductor foreign ownership ratio)

A single-session gain of +0.23pp is the largest in the sample. A further ≥ +0.10pp next session confirms the equipment-side handoff; 0 to +0.10pp indicates decaying momentum; negative ends this rotation leg.

Whole-thesis void condition: if any one of the three semiconductor names shows foreign net selling, the "rotation along the semiconductor line" thesis is void in its entirety, with no partial retention.

7. Methodology and reproducibility


Public-layer market research based on public market data and proprietary calculation methods. Not investment advice. Data basis is stated at the top; all calculations are independently reproducible using the conventions in Section 7.

想要更深入的期权数据与实盘分析?了解 AlphaGBM →